选择岭回归参数k值的改进方法及其在农药定量构效关系中的应用

提出改进的预报相对误差法选择岭回归参数ｋ，以平均预报相对误差替代预报残差平方和，并抑制过拟合。该方法应用于苯乙酰胺类除草农药定量构效关系的二次建模，效果良好，预报精度优于残差平方和方法。

An Improved k-value Selection Method in Ridge Regression and Its Application to Quantitative Structure- Activity Relationships of Herbicides

A new criterion for k-value selection, the key problem existed in ridge regression, was proposed based on the combination of the average predictive relative error with average self-checking relative error instead of the sum of predictive residual squared error. By applying this new method, the results indicated that the overfitting was debased and the model stabilization was improved at the same time. Therefore, it can be successfully used to the quantitative structure-activity relationship simulation of pesticides.